Reproducibility and robustness of economics and political science research

The study examines the reproducibility and robustness of 110 articles published in leading journals in economics and political science by reproducing the original results using the same data and code and testing how sensitive the findings are to alternative analytical choices. The results show that about 85% of published findings are reproducible, but only 72% of statistically significant results are robust, meaning they remain significant and in the same direction under reanalysis. Effect sizes are on average very similar to the originals (the median is about 99% of the original effect), but results tend to become somewhat weaker upon reanalysis, suggesting the presence of some publication bias.